The results show that the feature should be used for tables, where latch convoys already happen or are about to happen. In a not-so-heavy workloads, it brings a small overhead. You can see that inserts with 100 parallel sessions are 16% slower for a table with the optimized sequential key. However, when you hit the convoy, the feature definitely helps.
This graf is critical: if you don’t have high enough concurrency on the table, insertion can be a little slower than otherwise, so don’t go setting this for every table.